Why AI is a Game Changer for Pharmacovigilance and Safety

Contributed Commentary by Bruce Palsulich

July 7, 2021 | When new drugs, vaccines, and medical devices are developed and enter the market, one of the most important things to address is safety—the tracking of adverse events (AEs) related to the usage of medicinal products. According to Ernst & Young, large pharma companies currently have to deal with an average of 700,000 adverse event cases a year. But that number is rising quickly, with IDC estimating that caseloads are increasing by 30-50% annually. And, with COVID-19 vaccines having entered the market, that number is estimated to increase to more than one million a year for some companies.

As life science organizations try to gather and analyze the explosion of new data related to adverse events, it is becoming increasingly difficult to convert that information into meaningful conclusions. Further, companies must sort through which adverse events need immediate attention and which are lower priority. For example, an adverse event case where a patient is hospitalized is considered a higher priority than one where a patient reports a non-serious headache. Additionally, aggregated case data gives a more holistic safety picture than any single case, and this aggregate data is the basis for signal detection, i.e. mining the data for patterns that can indicate previously unknown risks of a product.

The secret weapon in this data battle is automation and artificial intelligence (AI). A well-designed, automated system using the right technology can eliminate tedious and repetitive tasks, reduce data entry errors, and even enable “predictive” signal detection. In fact, according to Ernst and Young, companies could reduce their pharmacovigilance costs by nearly 45% by automating manual steps.

From Challenge to Opportunity

Although the life sciences industry can understandably be risk-averse in the face of new technologies, it’s only a matter of time until AI becomes a commonplace solution in the day-to-day workings of every safety program. By applying AI and data science approaches, companies can turn the overabundance of data from a challenge to solve into an opportunity to leverage.

Life science organizations around the world are starting to see the real-world benefits that AI technology can bring to their own operations. Not only will AI bring speed, efficiency, and accuracy to safety programs, but it will also help practitioners focus on what they do best—evaluating the safety and effectiveness of their medicinal products—instead of being consumed by administrative tasks.

Focusing on adverse event case intake, AI can be applied to a wide range of data types, from forms with a defined structure to images. It is also possible to extract and analyze data from unstructured sources like journal articles or emails. Once the documents have been automatically structured and processed, they can be separated into “routine” cases that can be handled entirely by software, and “high-priority” cases that require review by safety specialists.

The insights provided by AI also enable safety evaluators to make more informed observations; for example, with new techniques such as neural signal detection, multimodal signal detection, and predictive signal detection. AI can assist these observations by integrating disparate big data sources, which could include traditionally collected post-market adverse event reports combined with electronic medical records (EMRs), journal articles, and information on chemical structures, pharmacodynamics, pharmacokinetics, and genomics.

Such insights help specialists make more informed benefit-risk evaluations, cross-checking patterns of data to better understand the safety profiles of their products, and extend the boundaries of scientific research. Pharmacovigilance and safety aren’t just about identifying serious risks—it’s really about honing suitability. By leveraging the analytical capabilities of AI, experts can better determine which patient populations (cohorts) respond well to certain medications and which ones don’t. This helps pharmaceutical companies avoid a market recall when the therapies could continue to benefit the lives of patients responding without adverse events. 

Gazing To The Future

Today, AI is best seen as a technology for augmenting human work. While “touchless case processing” may be possible in the future, studies have shown that the combination of AI and human intervention provides for a much lower rate of error than either one by itself. A 2016 White House report on AI studied this augmented approach for the examination of radiological images and concluded that the machine had an error rate of 7.5% while humans alone had a 3.5% error rate. Combined, the error rate dropped to 0.5%.

There are certainly aspects of safety case processing that are currently more suitable for greater degrees of touchless automation. For example, products that have been in the market for a considerable period of time and are well understood naturally require less human intervention because the safety profiles are well-known. Non-serious cases, consumer cases, and patient support programs may also be candidates for touchless processing that can be implemented within the next few years.

However, as the capabilities of various systems improve over time, more widespread use of touchless processing will become feasible. Given that likely future, companies can begin now to plan for and lay the groundwork for a touchless case processing system. The key is to adopt a stepwise approach, implementing and validating each automation area one at a time, to build an end-to-end automated process building incremental confidence and trust.

As capabilities improve over time, a phased approach of introducing automation and AI into the pharmacovigilance process can pay significant dividends. As companies and safety teams become more comfortable with automation and adjust workflows accordingly, they will gain the knowledge and confidence to increase levels of automation.

Companies can prepare now for this shift toward more end-to-end use of AI, by having the right technology in place, and teams trained and ready. While there are challenges ahead, including how to align with regulatory agencies, companies that start now will benefit from AI sooner and improve their safety management in a highly competitive market.


Bruce Palsulich is VP of Safety Product Strategy, Oracle Health Sciences. He leads safety product strategy for Oracle Health Sciences. This portfolio includes Argus Safety, the industry leading adverse event case processing and analytics solution, and Empirica Signal, the standard for signal detection and risk management. Bruce has more than 30 years of experience in the Healthcare and Life Science industry including 25 years in pharmacovigilance. He can be reached at bruce.palsulich@oracle.com.